🚀 FusionVis Platform

An Intelligent Document Understanding Platform for Education and Academic Scenarios

Lecture Parsing · Academic OCR · Knowledge Extraction · Intelligent Interaction · Structured LaTeX Output

Simplified Chinese Version

---

Project Introduction

This project is part of the 2025–2026 National College Student Innovation and Entrepreneurship Training Program (China)
Developed by a student team from the School of Data Science and Engineering, East China Normal University

FusionVis Platform is an intelligent platform focused on educational learning assistance and academic document understanding, integrating:

• Lecture and courseware parsing
• Academic paper OCR
• Structured knowledge extraction
• Frontend-backend interactive systems
• Local model inference capabilities

The platform is suitable for:

• Educational learning assistance
• Academic material organization
• Research-oriented knowledge management
• Intelligent PDF / PPT parsing
• AI document understanding experiments

Software Copyright

China Software Copyright: FusionVis: A One-Stop Platform for Learning Document Processing [Short Name: FusionVis] V1.0

Registration number: 2026SR0633721

The copyright in China is owned by East China Normal University.

Team Members

Yu Ke GitHub

Kenan Wang GitHub

Zhicheng Xu GitHub

Yumeng Yan GitHub

Chuhe Zhang GitHub

---

Platform Showcase

System Interface

Educational Learning Mode

Academic OCR Mode

Lightweight Chrome Extension

---

Project Highlights

Dual-Scenario Integration Simultaneously supports: Educational courseware learning Academic document analysis Providing unified access for both classroom and research scenarios.	Frontend-Backend Separation Supports: Independent deployment Distributed scalability Microservice-based architecture
Local Model Support Maintains local model directories for: Offline deployment Private environments GPU-accelerated inference	High Extensibility Supports rapid replacement of: OCR models Inference pipelines Frontend modules Knowledge enhancement components
Deeply Optimized Nougat OCR The platform adopts a fine-tuned version of the open-source Nougat model, further optimized for: Academic formula recognition PDF structure restoration Stable LaTeX output Complex page parsing In addition, OCR outputs are refined through LLM-based logical correction and structural verification, further improving: Mathematical formula recovery accuracy Consistency of tables and document structures Stability of LaTeX generation The model training methodology and related files will be gradually released in future publications and open-source plans.	LLM Verification and Knowledge Enhancement The lecture parsing and intelligent question-generation modules incorporate: LLM result verification mechanisms Hallucination suppression pipelines Multi-stage content consistency checks to reduce the risk of incorrect knowledge generated by large language models. The platform also integrates extensive knowledge bases and enhancement components, including: RAG-based retrieval augmentation Vector databases Multi-source knowledge expansion Cross-domain academic knowledge association Ensuring: Content completeness Continuous knowledge updates Academic information reliability Stable educational analysis performance

Project Structure

FusionVis/
│
├── backend/
│   ├── smartlec/        # Educational learning backend
│   └── nougat/          # Academic OCR backend
│
├── frontend/
│   └── web/             # Web frontend
│
├── models/              # Local model weights
├── storage/             # Runtime outputs and cache
├── docs/                # Supplementary documentation
└── figures/             # README display images

---

Quick Start

Environment Setup

Recommended environment:

Environment	Version
Python	3.11+
Node.js	18+
npm	Latest
Docker	Optional
GPU CUDA	Optional

---

Start the Educational Backend

Navigate to:

cd backend/smartlec

Run:

python -m venv .venv
.venv\Scripts\activate

pip install -r requirements.txt

python -m app.main

Linux / macOS:

source .venv/bin/activate

---

Start the Academic OCR Backend

Navigate to:

cd backend/nougat

Run:

pip install -e .

python app.py

---

Start the Frontend

Navigate to:

cd frontend/web

Run:

npm install

npm run dev

---

Docker Deployment

Educational Backend

docker build \
-f backend/smartlec/Dockerfile \
-t fusionvis-smartlec \
./backend/smartlec

Academic OCR Backend

docker build \
-f backend/nougat/Dockerfile \
-t fusionvis-nougat \
./backend/nougat

Web Frontend

docker build \
-f frontend/web/Dockerfile \
-t fusionvis-web \
./frontend/web

---

Functional Modules

Module	Function
SmartLec	Lecture parsing and educational assistance
Nougat OCR	PDF OCR and structure restoration
Web Frontend	Unified interaction interface
Local Models	Local inference and model management
Storage Engine	Document caching and result management

---

Usage Scenarios

Educational Learning Mode

1. Start SmartLec and Web services
2. Upload lecture slides / PDFs
3. Obtain:
   • Knowledge extraction
   • Content summarization
   • Learning assistance
   • Intelligent parsing results

---

Academic Document Mode

1. Start the Nougat OCR service
2. Upload academic paper PDFs
3. Obtain:
   • OCR results
   • Structured documents
   • Academic reading assistance

---

Team Responsibilities

Member	Responsibility
Yu Ke	Project Lead / System Architecture / Model Research / Frontend Optimization
Kenan Wang	Backend Development / Frontend Development
Zhicheng Xu	Model Research / Extension Development
Yumeng Yan	OCR and Version Control
Chuhe Zhang	System Testing and Deployment

---

Development Roadmap

Completed

• Educational courseware parsing
• Academic document OCR
• Web frontend system
• Dockerized deployment
• Local model support
• Multimodal knowledge association
• RAG-based knowledge enhancement
• Vector database integration
• Multi-user system
• Cloud collaborative parsing
• Enhanced PDF structure restoration

---

Related Links

Technical References

• Docker
• Node.js
• PyTorch

---

Secondary Development Suggestions

Recommended additions:

• .env.example
• Docker Compose
• Automated deployment scripts
• Frontend environment variable templates
• Model auto-downloader
• CI/CD Workflow

---

Sponsor

If this project helps you, feel free to:

• Star
• Fork
• Submit PRs
• Share the project

---

License

MIT License

Made with ❤️ by FusionVis Team